This is primarily done by using ConcurrentHashMap, which has lock-free reads and thread-contention-based writes. Only one thread at a time can register threads, however, as baking events isn't thread safe (and there's no reason to make it thread-safe anyway).
My own benchmarks indicate 1.4-2.2 million operations/ms throughput and approximately ~1ns/event post for four threads posting events. (commit: 59208aa)
